Do tropical typhoons smash community ties? Theory and Evidence

Transcription

Do tropical typhoons smash community ties? Theory and
Evidence from Vietnam
Yanos Zylberberg∗
Paris School of Economics
Preliminary and incomplete draft
Abstract
In rural economies, risk-sharing arrangements through networks of relatives and
friends are common. Indeed, contract enforcement issues impede the development of
formal insurance mechanisms. However, after a disruption of the allocative instruments of the market, the prerequisites under which informal arrangements are feasible
might evolve. I rely on a model of imperfect commitment to derive predictions on the
sustainability of risk-sharing arrangements in the aftermath of extreme events. I then
test these predictions on a representative panel data in Vietnam, using tropical typhoons trails and wind structures. The estimation of a structural equation derived by
the theory is compatible with a model of imperfect commitment where the aftermath
of natural disasters is associated with stronger enforcement mechanisms at commune
level. Allowing for altruistic sentiments or coordination among the members of a
community, I find that a resurgence of charity or a higher level of cooperation explain
this unexpected result. The influence of pre-disaster social norms and existing ties to
prevent disruption of integrative mechanisms in the community gives support to this
interpretation. Finally, communities having already suffered important trauma show
greater signs of resilience.
Keywords: Natural disasters, informal insurance, coordination, imperfect commitment.
JEL classification: D85, O12, O17, Z13
∗
PhD Candidate, Paris School of Economics, 48 Boulevard Jourdan, 75014 Paris. Phone: (33)1 43 13
63 14. Email: [email protected]. I thank the Joint Typhoon Warning Center, the National Archives and
Records Administration, and the Centre for Research on the Epidemiology of Disasters for providing the
data. Special thanks go to Nicholas Minot, Barbara Coello, Clément Imbert for help with the VHLSS data
and Francesco Avvisati for aesthetic considerations.
I.
Introduction
The recent earthquake in Haïti has highlighted the risk of double penalty in the wake of a
severe shock: the disruption of the classical allocative mechanisms seems to be accompanied by a rise of certain anti-social behaviors, pointing out a potential destruction of social
links. In certain situations of despair, individual motives and uncertainty on the attitudes
of others might overwhelm the belief in social coordination. Observers in Haïti even fear
difficulties in restoring the initial social environment and revivify the pre-existing community ties. Fortunately, international assistance was granted on a large scale. Furthermore,
Haïti has benefited from the impact of a large number of migrants in the United States,
and from being a former French colony. As put forward by Eisensee & Strömberg [2007],
international assistance are driven by other parameters than immediate needs. The tropical cyclone that struck Myanmar during the spring 2008 left thousands of people killed
without an equivalent media mobilisation.
In rural Vietnam, formal institutions designed to insure against income fluctuations are
failing. For large shocks, the decentralization has led to much less coordinated responses
from regional authorities. The interventions of NGOs, firms or public organizations (external assistance in general) are not always correlated to the real losses and come often
with a penalizing delay. Development of long-term crops and specialization1 have not been
accompanied by the creation of formal insurance schemes or institutions designed to facilitate the recovery. Credit constraints and the absence of private insurance penetration
coupled with the weak prevention schemes rule out the possibility of any external response.
Jacoby [1993] and Kochar [1999] show some evidence in favor of additional family labour
supply in response to crop shocks. Off-farm employment seems to allow the household to
insure a part of the losses incurred by crop shocks, explaining partly the flat consumption curve compared to farm income. Savings, as Paxson [1992] highlights, is also a way
to disentangle consumption from transitory income. Nonetheless, these instruments suppose the existence of off-farm flexible employment, positive net wealth and liquid assets
or unconstrained credit markets. As a consequence, households in rural economies often
deprived of those tend to rely on informal networks and non-market instruments. The
role of migrants has been highlighted in many studies. Yang & Choi [2007], using rainfall
shocks in the Phillipines, show that income shocks are partially (60%) covered by foreign
remittances. Households with migrant members show a flat consumption path while consumption and income are strongly correlated in households without migrants. Overseas
remittances appeal usually to the generosity of household members (or considered as such),
donations or informal loans involve contracting parties outside of the primary or secondary
family circles. Non-market insurance for idiosyncratic shocks is often provided by subgroups (friends, relatives, colleagues, neighbors...) of the entire community. Whether this
insurance is motivated by charity or simply as a part of a community implicit contract, a
lot of anecdotal evidence has been built upon this community response as a complement
to intertemporal instruments of risk-sharing.
Rosenzweig [1988] and Coate & Ravallion [1989] are the seminal papers raising the
importance of implicit contracts and informal risk-sharing arrangements in rural areas.
Imperfect commitment substantially constrains informal transfer arrangements. As a consequence, the privileged network is the network of relatives. Family links are the inherited
part of the social networks a household might belong to. In other words, the cost for
creating these links are supposedly lower in the family sphere. The larger the sphere the
1
responding to market incentives.
2
higher the costs2 . Along the same lines, Foster & Rosenzweig [2001] points out the fact
that commitment is more credible in networks of relatives as monitoring should be tighter
than in networks of friends or neighbours. Care for the contracting parties’ welfare plays
an important role in ameliorating commitment constraints. Interdependent preferences
(Pollak [1976]) might indeed weaken the incentives to default on the implicit contract as a
default hurts both the future fluctuations of the household and the current income of the
other party. The main insurance mechanisms (zero-interest informal loans combined with
pure transfers) and the status of the contractor reported in the Philippino villages studied
by Fafchamps & Lund [2003] confirm that tighter monitoring weakens considerably commitment issues in risk-sharing networks. Trust and monitoring issues seem to prevail. As
occupational activity of friends and relatives are often close to the household’s, there might
exist an arbitrage between diversification and monitoring capacities. Fafchamps & Gubert
[2007] establish that, leaving aside friendship or kinship, geographic and social proximity
are the major determinants of mutual insurance links among villagers. Such agreements
should be useless since all members might be affected while the possibility of risk-sharing
mechanisms at village level may be questioned as they require a certain amount of trust,
coordination or altruism to be enforceable.
After severe shocks, risk-pooling could improve total welfare even more efficiently than
after small fluctuations of income. However, the scope of classical informal insurance networks relying on relatives and friends make them particularly vulnerable to geographically
and occupationally co-moving shocks, which is why households should not be able to fall
back on these insurance networks when hit by a natural disaster. Furthermore, implicit
contracts at a higher level (the village) might not be robust to large fluctuations as they
suppose coordination from the community to commit credibly and punish deviations from
the informal agreements. The failure of market allocative mechanism might then also be
accompanied by a failure of implicit contracts. Douty’s seminal work on natural disasters’ aftermath showing the resilience of ’community feeling’ after a large and unexpected
shock has recently found a counterpart in the economic literature. Bellows & Miguel [2006]
and Bellows & Miguel [2009] show that individuals whose households have been directly
affected by the 1991-2002 Sierra Leone civil war were more likely to show a "community
feeling", being more likely to participate in community meetings, join political groups.
The indicators used in this study differ certainly from altruism or trust presented as the
prerequisites for implementing risk-sharing arrangements within a network. However, correlations might exist if we think of these indicators as reflecting a more global feeling of
trust or social coordination. Douty [1972] describes the use of informal social networks in
the wake of severe environmental shocks as an unexpected pattern of behavior. The confusion and uncertainty in the aftermath of the shock should lead any agent to go into her
shell - or the sphere where actions are directly controlled by herself. Surprisingly enough,
Douty remarks that residents affected by a natural disaster are inclined to be more charitable toward other members of the community. The reason advanced in Douty’s article
is the following: a natural disaster destroys the classical allocative mechanisms (in rural
Vietnam, no real allocative mechanisms are contingent to the occurrence of a catastrophe),
market coordination can then not be assured. As a consequence, primary units (households) coordinate themselves ignoring market allocative mechanisms. Douty points out
the fact that disasters of a sufficient magnitude often creates a super-organization headed
by pre-disaster civic leaders at community level. Informal allocative instruments seem less
used in normal times as market institutions might be more efficient. An issue not tackled
2
the expected gains are also detailed in the ’social capital’ literature (Glaeser et al. [2002]).
3
in the literature is how coordination is made easier during bad times. The non-market
coordination evoked by Douty is what I will consider as informal interactions. In other
words, this paper will focus on every allocative mechanism between or within households
that does not rely on market coordination. Domestic and overseas remittances from migrants, donations or low-interest loans, relief funds, public labor are the main observable
components of the community response.
Relying on an extended model of imperfect commitment developed by Ligon et al.
[2002], I derive classical predictions on the evolution of informal transfers in risk-sharing
groups after the realizations of large income losses. I then study small deviations from
the base model allowing for blurred contingencies, sub-group deviations, new entrants and
coordination mechanisms. Using a representative panel data in Vietnam between 2004 and
2006 matched with typhoon trails and wind matrices, I find that risk-arrangements across
households in communities affected by a cyclone are unexpectedly more efficient than for
idiosyncratic shocks. The estimation of a structural equation derived by the theory is
compatible with a model of imperfect commitment with higher punishment threats at
commune level in the aftermath of disasters. I show that this increase in monitoring
capacities is associated with a resurgence of charity or cooperation, laying the foundations
for risk-sharing at a level where normal events are not assured. Pre-disaster social norms
and coordination affect more the level of informal transfers following a typhoon than after
some idiosyncratic shocks. More importantly, communities seem to build social capital
after a disaster as communes having suffered important trauma in the past show greater
signs of resilience.
To our knowledge, this paper is the first paper of this literature focusing on large and
correlated income fluctuations risks and identify a learning-pattern in the enforcement
of implicit contract within a group. This paper also makes interesting methodological
contributions by estimating directly the self-enforcement model and, at a different level,
by matching cyclone trails with microeconomic data.
We present in section II. theoretical predictions on the enforceability of informal contracts. Then, we discuss the strategies to construct a consistent dataset and describe the
magnitude of tropical typhoons as well as reliance on informal transfers for rural households
in section III.. The section IV. presents the empirical strategies to test the theoretical implications. Extended results are discussed in sections V. and VI., focusing on cooperation
and the importance of past traumas.
II.
A.
Theoretical model of contract enforcement, monitoring
issues and natural disasters
A benchmark
In this section, we assume that there are a fixed number of households i = 1, ..., n receiving
an income yi (s) depending on the state of nature s at period t. The shocks are memoryless
and follow then a Markov process3 . The utility function is strictly increasing, concave
and similar for the n households. We suppose that households are infinitely lived and
3
both assumptions (finite number of state of nature and history-independent shocks) are essential.
4
maximizes their expected utility:
max
∞
X
β τ −t u(ciτ )
τ =t
We suppose that saving and borrowing are impossible. We rule out the existence of smoothing instruments. We assume that sharing of resources is unconstrained in this group of
households. Any reallocation of resources is theoretically possible. In the following section, we will refer to a reallocation as an anti-symmetric transfer matrix T(n,n) with the
coefficient on the i-th row, j-th column being the net transfer received by the household i
from the household j. An allocation of resources can be represented by an anti-symmetric
transfer matrix, this representation will allow us to rely as closely as possible on the initial
model of imperfect commitment.
Nevertheless, legal contracts are not feasible, which imposes the presence of a punishment to ensure that implicit contract will not be systematically violated. I will consider
a punishment P i (s) without specifying if this punishment derives from reputation mechanisms, exclusion from other activities (marriage market...). In addition to this external
cost, households who deviate will be excluded from the risk-sharing arrangements from
then on. The exclusion threat is supposed credible. We discuss this hypothesis in the extensions. The transfers can be made contingent to the realization of the state s (considered
verifiable). Contracts (antisymmetric transfer matrices T depending on the state s) are
enforceable if the punishment for deviating is higher than the instantaneous gain. In other
words, at a period t and for the state st , the agent i has no interest in deviating in period
t as long as:
a
Ut,i (st , T ) ≥ Ut,i
(st ) − Pi (st )
where:

P
 Ut,i (st , T ) = u cit + k Ti,k (st ) + Et Vti (st+1 , T )|st

a (s ) =
Ut,i
t
P∞
τ =t β
τ −t E
t
i
u(cτ )|st
utility derived from the contract
utility derived from autarchy
Let us remark that the concavity of the utility function ensures that the set of enforceable contracts will be convex. Any combination of enforceable contracts will also satisfy
the sustainability constraints. This assumption guarantees that the Slater conditions are
verified as the no-transfer contract does not bind any of the enforcement constraint. The
solution of the method of Lagrange multipliers generalized to inequality constraints is thus
optimal following Karush-Kuhn-Tucker.
Over the set of enforceable contracts, the optimal contracts maximizes the expected
utility of agent i keeping then utilities of the other households
above the surplus expected
o
a
from violating the contract Ūj (st ) = Ut,i (st ) − Pj (st )
. The program can be written
j6=i
as follows for the household i:
Vti (st ) = maxT Uti (st , T )
 j
(λj,i )
 Ut (st , T ) ≥ Ūj (st ) ∀j
u.c.
 j
Vt (st+1 , T ) ≥ V̄j (st+1 ) ∀j, st+1 (βµj,i (st+1 )P (st+1 |st ))
(MC)
Let us define Λj,i as the matrix composed of the columns of shadow prices λj,i related to the
maximization of household i under the enforcement constraint induced by the household j
5
at the contracting period. Let us define Θj,i,st+1 as the matrix composed of the columns of
shadow prices µj,i (st+1 ) related to the maximization of household i under the enforcement
constraint induced by the household j at the following period contingent to the state st+1 .
Intuitively, these shadow prices are the change of utility of household i for the optimal
solution of this optimization problem when relaxing the constraint on the household j and
decreasing the transfer from j to i by one unit.
The second-best contract is the optimal contract in the set of enforceable contracts.
This contract can be represented by the following equations:
• the enforcement constraints at period t and at period t = 1 for all households:
 j
 Vt (st+1 , T ) ≥ V̄j (st+1 ) ∀j, st+1

Utj (st , T ) ≥ Ūj (st ) ∀j, st
• the envelope and the first-order conditions:
P
u yti + k Ti,k (st )
Λj,i (st ) = P
u ytj + k Tj,k (st )
∀j
P
0 y i + T (s ) +
T
u
i,j
t
t
k6=j i,k
Λj,i (st ) + Θi,i (st+1 )
= P
j
1 + Θj,i (st+1 )
u0 yt − Ti,j (st ) + k6=i Ti,k
∀j, st+1
The third and fourth set of equations corresponds to the first-order conditions. The first
and second set replicates the enforcement constraint at the following and initial periods.
I can define a matrix of bounds for the marginal ratios (as the ratio of marginal utilities
between i and j is strictly increasing in Ti,j ). The first set of equations specifies then that
marginal ratios between j and i can not fall below a certain threshold Λj,i (st+1 ). This set
of equations can be interpreted as a convex set Ω(st+1 ) in which the transfers between the
members of the risk-sharing group are enclosed. As long as the constraints for i and j does
not bind at period t + 1, the shadow prices Θj,i (st+1 ) will be equal to 0 and the marginal
ratios of utilities in t + 1 will be equal to the marginal ratios at t. If the constraint binds
for the household j in the state st+1 , the ratio λj,i at period t for i will be adjusted upward
in period t + 1 so as to satisfy the enforcement constraint for the household j.

if Λj,i (st ) > Λj,i (st+1 )
 Λj,i (st )
Λj,i (st+1 ) =

Λj,i (st+1 ) otherwise
As long as the transfers which would maintain the ratios of marginal utilities equal
across time do not violate the enforcement conditions, the ratios of marginal utilities are
kept constant. This contract ensures then perfect insurance among the contractors when
the punishment costs are high enough. Once a marginal ratio is too low, the household
unaffected by the catastrophe might be tempted to violate the terms of the implicit contract
as the payment might be too important. The optimal contract readjusts the targeted ratio
downward to the limit where the contract remains enforceable. This implies first that
current payment will be lower than what an unconstrained contract would specify to keep
the incentives intact. Second, this readjustment is permanent4 , the ratio adjust to the
4
before the next violation.
6
new target5 . An example of how ratios might evolve with potential incentives to deviate
is shown in figure F1 in the appendix
B.
Extensions
Cooperation
In the previous section, we have considered a general punishment function encompassing exclusion from informal markets, reputation issues... Tighter monitoring, contempt
from the community for disregarding the terms of the contract or contingent punishment
(exclusion from other activities) both alter and relax the enforcement constraints, leading
to contracts closer to the first-best contract. Let us suppose that there exists a fixed (independent of the state of nature) cost Pind for a household and a global cost P (st ) ∈ {0, 1}
resulting from a cooperative game of the risk-sharing group. Intuitively, the former can be
interpreted as guilt while the latter would be related to reputation issues and community
punishment. The game fixes at each period the level of punishment that will prevail if a
default occurs. We assume that the negotiation is memoryless: the level of punishment at
period t does not influence the level at period t + 1. The level of punishment depends on a
vote, we assume that there is a private cost c of being in the losing group. For tractability,
let us suppose that there is a continuum of households. The level of punishment is determined by the number of households voting for a community sanction (p : [0, 1] 7→ [0, 1]
the probability to have a sanction depending on the proportion of households who vote for
the sanction). This extension weighed down considerably the base model as the decision
to vote depends on the contract, whose enforceability itself is influenced by the punishment level. In order to simplify the analysis, we will focus on the enforceability of the
full-insurance contract Tf i . Suppose that with a community punishment cost equal to 1,
the full-insurance contract would be sustainable, in other words:
Λ(st , Tf i ) ∈ Ω(st ),
∀st
For each state of nature, we sort the households by the private gains expected from
the sustainability of the current contract. m = 0 corresponds to the most enthusiastic
household in favor of a sanction. The vote for the point of view of a single household has
an interest (except from the cost of not cooperating with the rest of the group) in the
situation where the least enthusiastic household could default or respect the agreement
depending on the level chosen by the community:
a
−Pind − 1 ≤ Vt,1 (st , Tf i ) − Vt,1
(st ) ≤ −Pind
Let us denote P ∗ (st ) the level for which the least enthusiastic household is indifferent
between the two options. Even when restricting ourselves to monotonic equilibria, there
always exists two Nash equilibria -the full sanction (FS) equilibrium and the full forgiving (FF) equilibrium . However, both equilibria are not simultaneously robust to the
addition of uncertainty on the set of information used by households to derive their best
response. Taking into account the scenario where households vote without any priors on
5
Incidentally, the only unaffected by an infectious disease in a village will pay less than what the firstbest contract would predict. Furthermore, to ensure sustainability, once she has paid her tribute to the
community, her targeted relative well-being will rise, the distribution will be in her favor from then on,
relatively to the initial contracting conditions.
7
others’ behaviors, the only robust equilibrium will be determined by the type of the ’pivotal’ household, m = p−1 (P ∗ (st )). A vote for the sanction would generate the surplus
a (s ). As a
Vt,m (st , Tf i ) − (1 − p(m))c while a vote against would trigger −cp(m) + Vt,m
t
consequence,

a (s ) ⇒ coordination on the punishment
 Vt,m (st , Tf i ) − (1 − p(m)) c ≥ −cp(m) + Vt,m
t

a (s ) ⇒ coordination on the non-punishment
Vt,m (st , Tf i ) − (1 − p(m)) c < −cp(m) + Vt,m
t
As remarked above, a coordination on ’no sanction’ would provoke a default. The fullinsurance contract is thus enforceable as long as:

a

 Vt,m (st , Tf i ) − (1 − p(m)) c ≥ −cp(m) + Vt,m (st )
h
i

 m = p−1 (− mini Vt,i (st , Tf i ) − V a (st ) − Pind )
t,i
In this framework, an increase in the individual reputation costs would imply a smaller
community sanction to ensure sustainability. The pivotal household will consequently be
more in favor of a sanction than before. Nevertheless, the cost of voting for a sanction
might be higher as the threat represented by the proportion of households disagreeing
should be higher. This latter effect depends critically on the value of c and the influence of
the pivotal household on the result of the vote p0 (m). Lower c would make the reputation
effect negligible compared to the loosened enforcement constraint. Similarly, an increase
in the coordination mechanisms might have mitigated effects. The pivotal household will
remain the same (for a given state of nature), and the pressure on the household from the
community depends on the position of this household in the community. When a large
part of the community backs up the household in the decision to respect the contract, an
increase in the coordination mechanism will reinforce the sustainability. On the opposite,
if a large group is inclined to deviate, the increase in cooperation will induce easier contract
terminations.
Presence of sub-groups and piled layers of risk-sharing
In the previous contracting model, we adopt a dynamic framework without considering
the possibility that contracts might be renegotiated. Without any renegotiating costs, the
contract would be renegotiated at each period and the dynamic game would switch to a
simple repeated static game. Assuming that renegotiation costs are equivalent to violation
costs, any attempt to renegotiate is punished by the community as a violation of the initial
contract. This model supposes that commitment is possible, the community commits not
to renegotiate a contract with households reluctant to give some transfers during a period.
Nevertheless, since the punishment is a sunk cost, ex-post renegotiation is always optimal
when the network links are threatened. Commitment is thus necessary to ensure that the
economy does not revert to repeated static Nash equilibrium (which is a non-cooperative
equilibrium in the non-altruistic analysis).
Any enforceable contracts T1 and T2 within distinct subsets of households N1 and N2
can be represented as two contracts within the set N1 ∪ N2 (putting null transfer functions
T1 and T2 for households in (N1 ∪ N2 )/N1 and (N1 ∪ N2 )/N2 ) as long as:
∀i,
PiN1 ∪N2 (s) ≥
8
PiN1 (s) i ∈ N1
PiN2 (s) i ∈ N2
Conversely, under certain conditions, any enforceable contract within N1 ∪ N2 (’the optimal contract’) can be represented as a contract between the groups N1 and N2 and
risk-sharing arrangements within each group. As long as punishment within the group
N1 ∪ N2 , PiN1 ∪N2 (s), is lower or equal than the combination punishment within each group
and the punishment, the transfers implied by the global contract can be replicated consistently in each group. The intuition is the following: the transfer process follows two steps.
First, theP
total transfer between
P the sub-groups is decided on the basis of the aggregate
transfers i∈N2 ,j∈N1 Ti,j and i∈N1 ,j∈N2 Ti,j predicted in the optimal contract. Second,
the transfers are divided within the group so as to replicate the optimal contract. As such,
the utilities are the same as those which would have prevailed in the optimal contract.
Similarly, the utilities expected in autarchy are the same. The only constraint for which
the optimal contract is enforceable is then:
∀i,
PiN1 ∪N2 (s) ≤
PiN1 (s) + PN1 ,N2
PiN2 (s) + PN1 ,N2
i ∈ N1
i ∈ N2
where PN1 ,N2 is the externality a single default imposes on the other members of a subgroup.
This framework is thus compatible under certain conditions with a decomposition into
non-overlapping sub-groups with centralized transfers between sub-groups. Nevertheless,
our baseline specification is certainly not robust to specifications where deviations from
the entire sub-group do not imply punishment from households in the same sub-group.
As highlighted by Genicot & Ray [2003] and in a less direct way in Bloch et al. [2007],
the sustainability of a contract in a sub-group might endanger the stability of the entire
group. While the former derives directly this result using a simple and stylized model,
the latter could give an intuition shared by the sociological literature. Social capital and
stronger ties among groups of individuals can reinforce the pressure an individual might
endure if she deviates from the arrangement. On the other hand, it might also endanger
the whole group with joint deviations from a sub-network with strong ties. These results
are obtained without considering any competition between contractors, the deviation of a
sub-group leads to arrangements within the sub-group (sub-group autarchy). Deviations
are not followed by renegotiation with the deviants. To be more explicit, in a society with
rich farmers, poor farmers and storekeepers, an arrangement between farmers might break
if rich farmers are affected by a shock leaving poor farmers unaffected. The latters could
have the incentives to deviate jointly, support the contempt of rich farmers and negotiate
a contract with the storekeepers. This possibility prevents any risk-sharing arrangements
between sub-groups. This remark holds equally for individuals.
Influence of natural disasters - breaking the constraints
Another exciting feature of the model developed in Bloch et al. [2007] is the concept of
fragility. Stressful circumstances might tighten some enforcement constraints and a group
"obtains the opportunity to break away". As income exposure to large shocks are certainly
correlated in risk-sharing networks (Fafchamps & Gubert [2007] show that the groups are
principally composed of acquaintances with the same occupation and geographic location),
the networks in rural economies are not designed to share efficiently risk (especially large
risks). As a consequence, joint deviation from a sub-group following a natural disaster is to
be expected. This event can be observed in the dynamic framework described above with
two additional hypotheses. First, no contingency exists in the contract for rare events, the
states of nature covered by the contract are not exhaustive. If no contingencies exist, the
9
violation of a constraint is not anticipated and two different outcomes are possible: some
households deviate from the contract and a the initial contract become void leading to
renegotiation among remaining members of the risk-sharing group. The renegotiation can
also incorporate the sub-group, so as to avoid a sunk cost Pi .
Second, a contract might allow deviations of a certain sub-group contingent to certain
events or might accept the entire group dissolution. Technically, the group dissolution
is easy to implement as it does not change critically the enforcement constraint. Let us
consider a subset of states of nature Sc ⊂ S. The expected utility becomes then:
!
X
Uti (st , T ) = u cit +
Ti,k (st ) +Et Vti (st+1 , T )|st , st+1 ∈
/ Sc +Et Vti (st+1 , T ∗ )|st , st+1 ∈ Sc
k
The maximization program is similar, given the renegotiated contract T ∗ :
Vti (st , T ∗ ) = maxT Uti (st , T )
 j
(λj,i )
 Ut (st , T ) ≥ Ūj (st ) ∀j
u.c.
 j
Vt (st+1 , T ) ≥ V̄j (st+1 ) ∀j, st+1 µj,i,st+1 P (st+1 |st )
(MSC)
Naturally, denoting T̃ (.) the solution of the previous program, the following condition
should hold:
T ∗ (.) = T̃ (.)
The results are similar to the results of the base model. With independent realizations
of s, the only difference would come from a modified discounted value. However, in the
general case, this discounted value depends on the current state st .
The two hypotheses generating an effective breach of enforcement constraints predict
dissimilar post-disaster risk-sharing. When unexpected, a shock could lead to voluntary
departure of a sub-group of individuals. The remaining individuals decide on a forcible
eviction of this sub-group. Depending on the model assumption, risk-sharing might still be
possible and a contract may be renegotiated among ’traitors’. The first set of hypotheses
predicts a breach in the community. Both sets induce lower level of immediate transfers
between contracting parties than what would have been designed by a non-enforceable
contract. The renegotiating parties consider the post-disaster income path without taking
into consideration the pre-disaster income path.
C.
Predictions derived from the theoretical discussion
a. Informal contracts could exist even in communities without any institutionalized punishment mechanisms. Unconstrained transfers are made so that the marginal ratios
of utilities between the members of the group are kept constant. Some ratios might
decrease or increase so as to ensure the contract sustainability.
b. When punishments are independent of the state of nature, a uniform increase in the
deviation cost extend the ’sustainable’ set. Local increase might have mitigated effects
depending on the role of the household concerned by the tighter monitoring mechanisms.
The presence of sub-groups with correlated incentives to default decreases the expected
punishment and the potential enforceability of a contract. Entrants play the same role.
c. With punishment mechanisms independent of state of nature, whether unanticipated or
part of a deviation clause, natural disasters should generate smaller instantaneous risksharing. They should not imply different transfers than other shocks of similar amplitude
10
if perfectly anticipated and not contingent-specific. When predicted or forgiven, postdisaster risk-sharing should not be smaller than before on longer horizons with the same
monitoring facilities. On the opposite, unanticipated deviations should imply exclusions
from the risk-sharing groups and lower aggregate risk-sharing on the long term.
d. With cooperation, the effect of higher coordination among the households might have
mitigated effects depending on the proportion of households willing to deviate. A higher
level of cooperation should trigger a higher level of risk-pooling in communities where
the coalition favors contract enforcement in the aftermath of a typhoon.
III.
A.
Description of the data
Construction of the datasets
Vietnam Household Living Standards Surveys
In this article, we use the Vietnam Household Living Standards Surveys which were carried out in 2002, 2004 and 2006 by the General Statistics Office. These surveys reproduced
quite faithfully a first wave of surveys (VLSS 1992/1993 and 1997/1998) organized with a
tight monitoring of the World Bank but departs from them by including an expenditure
module to the initial questionnaire. A panel is conducted between the three waves of 2002,
2004 and 2006 and the structure of the questionnaire remains stable. As shown in figure 1,
a very large number of districts are represented in this study and geographical indicators
are sufficiently precise to locate each commune6 in a district despite changes in enumeration
areas between 2002 and 2006. In each commune, 3 households are randomly interviewed.
The surveys we use contain a household survey and a commune survey. The former covers
household characteristics, education, health, housing conditions, employment, agriculture,
expenditure, remittances, ’social-oriented’ expenditures and credit access for each household while the latter focuses mostly on general living standards and, in particular, eligibility
to reforms, natural disasters and potential relief, agriculture and credit barriers at commune level. Investment in social capital as described in the introduction can be precisely
controlled with the expenditure module. Gifts, donations, investment in funds or inflows
such as domestic remittances are well documented. Controls for potential externalities,
financial constraints are present in the database. Unfortunately, the questionnaire is not
as detailed as General Social Surveys concerning membership in social groups, church attendance and indicators of trust. It is impossible to define precisely risk-sharing partners
or potential partners and reconstitute the friends and relatives networks. Furthermore,
the study have been conducted during several months (mostly during two periods, June
and September), generating difficulties when determining the relative exposure to a certain
event occurring contemporary to the survey.
Cyclone tracks and exposure to natural disasters
From Joint Typhoon Warning Center, I extract best tracks of tropical typhoons between
1997 and 2006 having landed or generated torrential rains on Vietnamese coasts. Wind
intensity, pressure, precise location, form and size of the eye are precisely documented
every 6 hours. This allows us to reconstruct the trails and the wind structure. I then
6
a commune is often composed of several small villages and represents one of the smallest enumeration
areas in Vietnam.
11
consider the potential average dissipated energy per km2 along the path of the cyclone for
each of the 600 districts composing Vietnam. The figure 1 shows the total wind structure
of a selected panel of cyclones (Vicente, Damrey and Chanthu) and an index of the local
exposure to tropical typhoons. In order to account for the floods associated to tropical
typhoons, I create a band7 along the path of the cyclone. As a consequence, I associate a
measure of wind intensity and an exposure to torrential rains for each cyclone. To control
for the potential exposure to such events, I use the Global Cyclone Hazard Frequency and
Distribution data and assess precisely the exposure profile of a district8 .
At district level, I compute the precipitation for each month between 1997 and 2002 (extracted from Global Energy and Water Cycle Experiment and the specific Asian Monsoon
Experiment which was an international collaborative research project as part of World Climate Research Program). I consider the average monthly precipitation level during these
six years and generate indicators of droughts and inundations.
Figure 1: Location of surveyed households in the panel between 2004 and 2006. Potential exposure to the passage of typhoons and 3 occurences: Vicente, Damrey (2005) and
Chanthu (late 2004)
Bombing intensity in Vietnam and Laos between 1965 and 1975
Combat Air Activities (CACTA) and Combat Naval Gunfire (CONGA) files, which
are both records of the U.S. Joint Chiefs of Staff, are provided by the Department of
Defense and the National Archives. Each entry documents an assault, giving information
on the type of ammunitions used, the aircraft, the alleged objective and an estimate of the
outcome. I use the total weight of bombs used during the attack as a proxy for the energy
dissipated around the impacted zone. I then interpolate these observations with the 600
7
whose width depends on the pressure reported by JTWC.
the data associates the exposure profile computed between 1980 and 2000 for ’squares’ whose dimensions are roughly 0.25 degree of latitude and 0.25 degree of longitude.
8
12
districts so as to create the total exposure to bombing between 1965 and 1975 for each
district. The figure presents the results showing the large spectrum of regions affected by
during the 1965-1975 window.
B.
Descriptive statistics
Magnitude of natural disasters
The figure 1 shows the geographic dispersion of surveyed households. Our surveys
covers almost 600 districts, and between 1 and 36 households surveyed by districts. A
first random draw determines the communes in the districts which will receive the visit
of a surveyor and another draw defines the sub-sample of households to be interviewed
in the commune. 75%9 of the surveyed households live in rural areas, 53% of those are
located in the Delta, 7% in coastal areas, 7% in hilly lands and 33% in mountainous areas.
As we construct the natural disasters exposure not on direct measure of income losses,
we compute in this section a rough estimation of the influence of each tropical typhoon
considered in this study. Relying on objective measures on exposure, we compute the
average exposure per commune for each cyclone. Using the weights provided by VHLSS,
this measure can be extended and provide an estimation of direct and indirect damages at
country level. We can then compare the results we find with estimations of direct damages
recorded in the EM-DAT10 database. Unsurprisingly, our index differs from EM-DAT
estimations. While EM-DAT announced approximately USD 900 millions of losses due to
the tropical typhoons between 2004 and 2006 and 300 millions for the typhoons that belong
entirely to the surveyed window11 , our measure predicts USD 580 millions of losses over the
surveyed window, approximately 1% of the Gross Domestic Product of Vietnam in 2005.
Besides the potential measure error implied by our estimation process, this difference can
be easily explained as EM-DAT provides direct estimations, indirect effects for typhoons in
and out of the surveyed window are not taken into account. Our index tends to highlight
also indirect and long-term effects. Even though none of the tropical typhoons studied
here were considered dreadful, economic damages in the aftermath of the shocks remain
significant. Being in the eye of Damrey12 presumably affected durably a whole region, a
dozen of districts present predicted losses over the year above 20% of their usual predicted
annual income. Districts of the central regions affected successively by Chanthu in 2004,
Kai-tak in 2005 and Angsane in the late 2006 underwent similar losses.
Failure of formal institutions
Private insurance is almost absent in our sample. Thus, only 5, 6% of the surveyed
households in 2004 have a formal non-life and not health-centered insurance contract and
less than 4, 5% when ruling out urban areas. The figures are similar for life insurance
contracts13 while health insurance seems to be more frequent14 but covers extremely small
amounts. 47% of rural households are currently reimbursing a loan. The proportion falls
9
The figures in this section are extracted from the 2004 wave. Descriptive statistics do not change
dramatically with other waves.
10
EM-DAT: The OFDA/CRED International Disaster Database (www.emdat.be), Université Catholique
de Louvain.
11
Xangsane having occurred in September 2006, only few households surveyed in October reported losses.
12
which provoked USD 220 millions direct damages according to EM-DAT.
13
respectively 5, 4% and 3, 8%.
14
respectively 39% and 35%.
13
respectively to 30% when we only consider loans contracted with a formal credit institution
(private or public credit institutions). Several households are reimbursing more than a
single loan but second and third loans are mainly informal. The interest rate per week
is roughly 1% for all formal credit institutions, which is extremely high. Long-term loans
(above 2 years) are rare. Loans for other reasons than durable investments represent
less than 10% of loans provided by formal institutions and 40% of those contracted with
friends, relatives. The access to formal loans seems to be restricted and does not respond
to consumption needs but to capital investments. However, the access to private insurance
and credit institutions do not account for the whole ’formal sector’ when it comes to the
analysis of non idiosyncratic risk.
We include state/regional intervention and NGO’s relief aid as part of the formal response to natural disasters. Indeed, these amounts are essentially destined to the commune
and are used to reconstruct roads and other public goods. The fact that the relief aid is
often dealt by the commune leader mitigates the reach of intervention of any single household. Using the commune questionnaire of VLSS and the amount and provider of relief aid,
we compute the correlations between these ex-post transfers and our measure of income
losses. The correlations are non significant at district level. Household-level correlation
between the aid declared by the respondent and income losses due to shocks is also not different from zero. Putting aside the reasons behind the relief aid, the amount are extremely
small. Allowance for disaster recovery hardly reaches 1% of the household income in the
most affected districts. Similarly, support from organizations at commune-level represent
more than 1% of the income in 2 districts only and a dozen of communes. To conclude,
the presence of ex-post transfers organized by regional or national authorities seems to be
correlated with the institutional environment more than immediate needs. Finally, credit
market access, private insurance penetration and external ex-post transfers (presence of
NGOs) are significantly correlated at district-level. Some communes benefit from the whole
range of instruments but a large proportion of communes (essentially in rural zones) are
deprived of the whole panel of risk-sharing mechanisms.
Informal risk-sharing arrangements
Informal risk-sharing arrangements are present in our surveys in the forms of gifts,
transfers, remittances and loans. In-kind transfers and payments are also collected. These
data are total inflows and outflows over the past year, except for the loan section for
which each transaction is recorded with the partner ’type’ (the partner can not be exactly
identified and the probability to have a partner in the study is extremely low). Gift
and donations are largely present in rural and urban areas. Only 10% of households
in rural areas and 20% of urban households had zero outflows during the past year. The
average outflow represents approximately 3, 5% of the income of each household. To confirm
these data on outflows, inflows of domestic remittances are absent for only a sixth of
the total sample. The average and median amounts received during the past year from
relatives and friends are respectively 10% and 2, 9% of the receiver’s income. The average
amount is biased upward compared to the outflows data and median amount reflecting
that a large number of households have relied on remittances only during the past year.
Unsurprisingly, the foreign remittances concern a much smaller part of the population.
11, 5% and 4, 5% of the urban and rural samples receive positive foreign remittances. In
line with the intuition that foreign migrants support financially aging households, the
average amount when present is six time higher than the average domestic remittances. It
represents approximately a third of the total income perceived by the domestic household.
These households are more urban, older and less active than the average household receiving
14
Table 1: Descriptive statistics on formal and informal instruments
rural
general
Income (USD 2004)
1382
Delta
.53
Hills and mountains
.37
Coastal areas
.10
urban
2511
-
presence of formal instruments
Life insurance
.04
.10
Health insurance
.35
.52
Non-life insurance
.05
.09
Formal loans
.30
.22
Loans for non-durable .02
.02
presence of informal
Foreign remittances
Domestic remittances
Informal loans
Zero-rate loans
Loans for non-durable
instruments
.04
.11
.83
.84
.14
.12
.11
.10
.04
.04
domestic remittances. They should also be less exposed to natural disasters in line with
these results. The loan part of the survey confirm the importance of informal transfers.
Thus, 15% have contributed to a revolving group or lent to another household in the
past year. This proportion is homogenous between rural and urban areas. The median
contribution represents 5, 6% of the lending family income.
Confirming these results, 10%15 of the surveyed households have borrowed the past
year from friends and relatives. An additional 4% are contracted between individuals not
declared as friends or relatives16 . The median ongoing informal loan represents 20% of the
borrowing family income (15% for the informal loans outside of the friends network, which
is comparable to the share represented by the formal loan face value compared to the income of the borrowing family). Informal loans interest rate show explicitly more altruistic
behaviors or easier enforcement/monitoring mechanisms when the relationship between the
contractors is tighter. Thus, interest rate required by relatives or friends is zero for 82%
of the rural household. This proportion falls to 12% when the contractor is an individual
out of the primary sphere and less than 2% when the contractor is a formal institution.
Surprisingly, informal loans do not appear to be more present in poor communities and the
proportion of zero interest loans are not higher in these poor communities. Similarly, the
presence of preferential credit has no influence on the whole community. Only households
actually benefitting from lower interest rates borrow more. Since they have a preferential access to credit, households rely less on informal loans and when they do, they obtain
milder conditions from their friends (94% of zero interest loans against 83% for non-eligible
households, reflecting the better outside option). Recent movers seem to rely a bit less on
friends. This statistics is confirmed at commune-level: large turnover is associated with
15
The following statistics are extracted from the subsample of surveyed families living in rural areas.
the status of this individual is unknown. In practice, she could be a retailer or a colleague but also a
usurer offering extremely high interest rates.
16
15
smaller importance of informal borrowing (friends, relatives or individuals of the community). The purposes of the loans differ significantly had it been contracted with formal
institutions or individuals. 80% of formal loans respond to long-term investments (durable
goods, capital, housing issues...) while the proportion hardly reaches 50% for informal
arrangements (mainly driven by housing issues). These arrangements are privileged for
consumption, medical issues, the maturity of these loans is thus slightly lower than the
average maturity.
Descriptive statistics on households affected by a typhoon
As shown in the following table, the evolution of some key variables in districts affected
by a typhoon confirms the intuition developed above. First, the results confirm that predicted losses at commune level (depending on the exposure of a commune - proportion of
assets exposed to typhoons) do find a counterpart at individual level! Affected households
have a higher probability of having contracted informal loans and lower probabilities for
formal loans. They have received higher domestic remittances but similar levels of foreign
remittances, insurance reimbursement or aid from NGOs. The outflows also documents a
higher activity for informal mechanisms following a disaster. Contribution to funds (including natural disaster funds) and lending are positively correlated to the predicted income
losses incurred by a disaster.
Table 2: Descriptive statistics on households affected by tropical typhoons
Direct consequences
on ’social’ inflows and outflows in 2006
Coefficient
(Standard error)
Income
Income in rural areas
Commune income (district FE)
Commune income in rural areas (district FE)
-.218
-.172
-.793
-.912
(.041)∗∗
(.047)∗∗
(.195)∗∗
(.231)∗∗
inflows
Probability to contract formal loans
Probability to contract informal loans
Domestic remittances and presents
Foreign remittances
Aid
Insurance
-.332
.359
1.22
-.210
.007
.262
(.162)∗
(.176)∗
(.296)∗∗
(.265)
(.193)
(.218)
-.394
.675
.893
-.562
.776
(.371)
(.261)∗
(.450)∗
(.301)†
(.310)∗
income
outflows
Donations and support
Contribution to funds
Entertainment
Funeral and death anniversaries
Lending
These are the elasticities to the average (within a commune) income losses induced by
natural disasters. Each regression includes the explained variable in 2004 and controls
for the expected exposure to natural disasters. Significances are indicated at 10%† ,
5%∗ , 1%∗∗ .
16
IV.
Empirical strategies and first results
The model developed in the previous section predicted a relationship between the level
of informal transfers, the history of transfers (establishing the conditions of the ongoing
contract), the revenue of the individual and the revenue of the rest of the group. Let us
suppose that the expected income in a sub-group is homogeneous and that the shocks are
small compared to the income level. We then linearize the equations, resulting in a linear
expression of transfers as a function of individual shocks and the historical λt (conditions
under which the contract has been designed). We denote ȳ i the expected component
of income, zti the unexpected component, τtj the aggregate net transfers received by the
household. Inflows are associated with positive τ ’s and outflows with a negative τ ’s. The
expression of the transfer function can be derived from
λi,j
s,t =
0
00
i + τ i)
u (y¯i ) + u (y¯i )(zs,t
t
0 ¯j
00 ¯j
j
u (y ) + u (y )(z + τ j )
s,t
∀i, j
t
As a consequence,
j
zs,t
+ τtj =
n
X
j
[zs,t
+
τtj ]
=
n
X
u
(ȳtj )
00
u (ȳti )
0
− u (ȳtj ) +
λi,j
s,t
" 0
u (ȳti )
1
00
#
0
u (ȳti )
u00 (ȳtj )
j=1
j=1
"
1
λi,j
s,t
00
j
λi,j
s,t u (ȳt )
#
−u
0
(ȳtj )
+
n
X
i
[zs,t
+ τti ]
00
u (ȳti )
i,j 00 j
j=1 λs,t u (ȳt )
i
[zs,t
+ τti ]
As the sum of transfers should be zero,
τti
=
i
−zs,t
n
1 X j
+ i
z −
Nt j=1 s,t
where Nti is defined as:
Nti =
0
u (ȳti )
ȳti u00 (ȳti )
!
ȳti
n
1 X
+ i
Nt j=1
n
X
u (ȳti )
j=1
00
j
λi,j
s,t u (ȳt )
0
u (ȳti )
ȳti u00 (ȳtj )
!
ȳti
00
Under a non-binding specification17 , we can notice that the last terms cancel out,
leaving a simple and tractable equation; the λ’s are independent of the state of nature and
N can simply be written as a function of the risk aversions and expected incomes of the
households::

Pn
j
i + 1

τti = −(1 − N1i )zs,t
i

j6=1 zs,t
N

t
t
(S1)
i j
P


 Nti = nj=1 ȳtjσ i
ȳt σ
With finite violation costs, there exists a threshold of transfers under which the household might be tempted to default and bear the violation costs. This threshold translates
into interdependent upper bounds for the marginal ratios18 when linearizing around the
17
suppose that the violation costs are almost infinite.
In other words, binding marginal ratios for a single household i with respect to the other households
are proportional.
18
17
expected income:
 i,j
i,k i,k
i,j

 λs,t γ = λs,t γ
∀j, k

 γ i,j = u0 (ȳ j ) ȳtj σi [z + τ ]
i
t ȳ i σ j i
t
The interpretation is straightforward. The closer a household is to the enforcement constraint, the less correlated the transfers will be. The agent with the strongest incentives
not to pay a transfer in a state of nature will be asked to pay a lesser amount than what
a first-best contract would have prescribed.
 i,j
i,j
i,j

λ
=
min
λ
,
λ
∀i
 s,t
t−1 s,t ,


ln λi,j
s,t
i,j
i,k
λs,t γ i,j = λs,t γ i,k ,
≤ min
ln(λi,j
t−1 ), ln
∀j, k
i,k
i,k γ
λs,t i,j
,
γ
∀i, j, k
(S2)
The equation S2 can also be interpreted as measuring the pressure exerted by external
households k on the transfers received by the household i. Thus, transfers destined to keep
the balance between incomes of i and j will be influenced by other households if and only if
both constraints (internal - between i and j - and external - between i and some household
k) are binding.
Empirical counterpart - Estimation of the equation S1

P j
i
i
i

 τt = −α∆zt + γ∆ j zt + εt
j
j
j
zt = yt − ỹt , ∀j

 ỹ j = f (y j , X j ), ∀j
t
t−1
t−1
The identification method relies on a three step process: first, we predict the level of
j
income in t, given observables Xt−1
in t−1. To perform this analysis, we do not impose any
structure on the model and construct bins grouping households with similar characteristics
in t − 1 (10 categories of income, age and education of the head, occupation, rural/urban
areas). The unexpected part is the difference between a household’s income and the income
of households sharing similar initial characteristics. Nevertheless, this income is unexpected
only for the statistician using a couple of observables and might reflect the graduation
of a young member of the household or the migration of another (which are certainly
expected and does not enter into insurance contracts). As a consequence, we extract this
unexpected part of income and decompose it into different kind of shocks. The third step is
the estimation of the transfer function using the shocks of the household and the potential
contractors.
Using typhoon trails, we identify the unexpected income component lost in the aftermath of a disaster. To derive this measure, we regress the unexpected income on two
indicators: the energy dissipated along the typhoon path and the probability of suffering
from heavy rains19 . This process generate a district natural exposure. Individual exposure
are determined using the assets owned by the households in 2004. To control for potential
bias induced by other regional factors, we control for the risk of being hit by a typhoon as
19
we construct a belt along the track whose width depends on the pressure at the center of the eye.
18
Table 3: Gifts and informal loans flows following natural disasters
Specifications
Specification (S1)
OLS
Coeff. (SE)
Own shock
Average shock on neighbors
(.08)∗∗
(.09)
-.318
-.10
Observations
District FE
Coeff. (SE)
-.286
-.055
6508
(.08)∗∗
(.10)
6508
Significances are indicated at 10%† , 5%∗ , 1%∗∗ .
predicted at the previous period (in 2004). The component of unexpected income predicted
by our natural manifestations is what we define as the natural shocks. Similarly, we identify
compulsory health expenditure shocks. We use a government program providing hospital
fees reduction as an instrument to avoid a "reverse causality" bias: a positive shock on
income could give the opportunity for a certain household to benefit from costly health
infrastructure, which would have been prohibited without the substantial lift in income. In
addition to these health expenditure shocks, we construct an unexpected shock with a loss
profile closer to those generated by natural disasters. At the beginning of 2004, the Avian
influenza epizooty (H5N1) has generated heavy income losses for the households owning
livestock (especially poultries). The relief provided by regional and national authorities
has been far from fully covering for the total income loss. Copying our estimation process
for natural shocks, we consider communes where community leaders reported heavy losses
from the epizooty and create an indicator of commune exposure. Individual exposure are
determined using the livestocks owned by the households in 2004, distinguishing poultries
from other types of livestocks. Contrary to the typhoon exposure, we can not control for
expectations of households in 2004 for these epizooty shocks. It seems reasonable to think
that the expansion of H5N1 through South-east Asia was not predictable. Nevertheless,
being affected by the epizooty could reflect bad coordination at commune level.
Table 4: Transfers of assets following natural disasters
Specifications
Specification (S1)
OLS
Coeff. (SE)
Own shock
-.473
.608
Observations
(.28)†
(.34)†
6508
District FE
Coeff. (SE)
-.511
.547
(.30)†
(.39)
6508
The first estimation concerning natural shocks proves that informal arrangements still
play a role after severe shocks. Thus, approximately 30% of income losses incurred by
a tropical typhoon is insured through domestic remittances and informal loans. We can
decompose this effect into two significant effects: loans contracted with friends account
for 18% while gifts represent 13% of these informal flows. The results are robusts to
the addition of district-level fixed effects or household control variables (age, ethnicity,
19
education level of the head, number of old parents and young children...). Combined
with asset transfers, we can not reject the full-insurance hypothesis. The fact that the
coefficient for shocks affecting the rest of the community is not close to being the opposite
of the coefficient for the household income fluctuations implies that our model does not fully
capture the motives explaining the level of informal transfers. Potentially, it could reflect
that data reject the hypothesis of homothetic invariance relatively to the amplitude shock.
In other words, risk-sharing might be more efficient in places where the catastrophe has been
particularly dreadful. Second, it could come from a bias linked to external interventions.
Domestic remittances are included in our measure of informal transfers and a part of the
transfers might occur outside of the districts. We discuss the importance of risk-pooling
across districts at the end of this section.
Table 5: Informal transfers and saving flows following non-natural disasters shocks
Informal transfers
Specification (S1)
OLS
Coeff. (SE)
Own shock
-.003
-.024
Observations
(.057)
(.069)
District FE
Coeff. (SE)
-.038
-.089
6771
Transfers of assets
OLS
Coeff. (SE)
Own shock
-.693
.972
Observations
(.207)∗∗
(.249)∗∗
6771
(.056)
(.071)
6771
District FE
Coeff. (SE)
-.592
1.16
(.212)∗∗
(.266)∗∗
6771
The correlations between informal flows and epizooty or health shocks are not significantly different from 0. The difference between these latter shocks and natural disasters
can be explained by:
• a difference in the profile of income losses incurred by the different shocks, health
shocks might be smaller in general. We should compare the degree of insurance of
health shocks to the degree of insurance provided by informal flows for the central part
of the loss distribution induced by tropical typhoons. With transaction costs, small
transactions might be absent and insurance is under-estimated as informal responses
to these individual shocks might have been censored. However, the epizooty triggered
a similar loss distribution at commune and individual levels and results do not differ
when considering only shocks related to the Avian influenza episode.
• a bias induced by the specification. Assuming non-binding constraints for the specification S1 and homothetic invariance as described earlier, we can only test if the
empirical results are consistent and predict constant λ’s.
• a specific feature of natural disasters implying a greater level of risk-pooling through
informal transfers despite similar responses for formal transfers.
20
Empirical counterpart - Estimation of the equation S2
We replicate the previous construction of expected income and unexpected shocks.
We aggregate the observations (we restrict the sample to the communes where exactly 3
households are interviewed) for each commune and create series of effective marginal ratios between the surveyed households, using a CARA utility function with a risk aversion
of 1.92. We then sort the 3 surveyed households of a certain commune along their predicted effective exposure to different shocks (tropical typhoons, health expenditures and
epizooty). We do not know if these households are actually members of a same risk-pooling
group or if they can consider themselves as potential partners. Nonetheless, these households (randomly chosen) provide a picture of the distribution of households in a commune.
Consequently, under the hypothesis that risk-pooling intervenes at commune level, the
supposed partners approach the profiles of real potential partners.
Having sorted the households such that binding constraints for the second household
imply monotonically binding constraints for the third, the marginal ratio between the
household representing the most affected households in the commune should be either
1,2
equal to the targeted marginal ratio λt−1
or to the transposed ratio imposed by the pressure
1,3
1,3
exerted by the least affected household λt γγ 1,2 . The intuition is just that this ratio between
the most affected household and the median household will be influenced by how much the
household standing for the least affected households is eager to compensate the resourceless.
Indeed, denoting BCi,j the event "‘the constraint is binding between i and j"’:
1,3
(BC1,2 = 1) ⇒ (BC1,3 = 1) ⇒ λt1,3 = λt
ln λ1,2
s,t =

1,2
 ln λt−1 if BC1,2 = 0

h
ln χ1,2
s,t
if BC1,2 = 1
i
1,3
1,3 γ
where χ1,2
s,t = λs,t γ 1,2 is the transposed ratio. The empirical counterpart can be written
as:

 ln λs,t = β̂ ln λt−1 + (1 − β̂) ln χs,t + εt if binding

ln λs,t = ζ ln λt−1 + (1 − ζ) ln χs,t + ηt
otherwise
where:
β̂ = P (BC = 0| ln λt−1 − ln χs,t ) = α + γ ẑt + νt
where ẑt is a normalized20 measure of the amplitude of the shock affecting the commune.
1,2
The system above is identified as long as Cov(νt , ln λ1,2
t−1 − ln χs,t ) = 0, i.e. conditional on
the difference between unconstrained and constrained λ’s, the unexplained weight given to
one or the other λ should not be correlated with this difference. In this framework,
• α and ζ are the respective probabilities that the constraint is effectively binding
conditional on the fact that the model respectively predicts it should be or not.
α > β is a test that the model predictions concerning potential default do coincide
with tighter constraints.
• γ is the elasticity to the amplitude of the shock of this propensity to be constrained
(when predicted). To grasp the intuition, γ is the additional weight of the effective
20
we use the hyperbolic tangent of the average effective exposure of the commune. Thus, an infinite loss
would be associated with a measure of −1.
21
enforcement constraint on the marginal ratios when the model predicts that the
enforcement constraint should be binding.
Table 6: Pressure on the enforcement constraints after a shock
Type of shocks
Specification (S2)
Natural shocks
Epizooty shocks
Specifications
Co.OLS
Un.OLS
Overweight induced
.642
(.334)∗
.202
(.023)
.798
(.023)
.938
(.335)∗∗
-.060
(.025)
.601
(.024)
.338
(.022)
.662
(.022)
.055
(.025)
.491
(.023)
Targeted ratio
Constrained ratio
Targeted ratio
Constrained ratio
Targeted ratios
Constrained ratios
Targeted ratios
Constrained ratios
Observations
.136
(.031)∗∗
-.136
(.031)∗∗
.072
(.047)
-.072
(.047)
Health shocks
Co.OLS Un.OLS
Binding constraint
-.868
-.591
(.199)∗∗ (.200)∗∗
.311
.075
(.021)
(.024)
.689
.546
(.021)
(.022)
Co.OLS
Un.OLS
-.090
(.043)∗
.308
(.021)
.692
(.021)
-.073
(.043)†
.016
(.023)
.488
(.022)
Non-binding constraint
.334
.067
(.022)
(.025)
.666
.476
(.022)
(.024)
.426
(.022)
.574
(.022)
.175
(.025)
.419
(.023)
Tests for equality (average commune shock)
.110
.023
.070
.117
(.033)∗∗ (.030)
(.032)∗
(.029)∗∗
-.116
-.023
.008
-.117
(.036)∗∗ (.030)
(.035)
(.029)∗∗
Tests for
.016
(.049)
-.022
(.050)
equality (10% commune shock)
.126
.076
.110
(.029)∗∗ (.031)∗
(.036)∗∗
-.126
-.167
-.110
(.029)∗∗ (.034)∗∗ (.036)∗∗
1756
1724
.069
(.031)∗
-.159
(.033)∗∗
.129
(.038)∗∗
-.051
(.040)
1855
†
∗
∗∗
Significances are indicated for the important variables at 10% , 5% , 1% . We display the
results for constrained and unconstrained regressions. We perform a maximum likelihood
process to account for a general variance-covariance matrix. The tests compare the weight
attributed to the targeted and transposed ratios for a commune hitted by an ’average’ and
10% shocks.
The results obtained during the first battery of tests are confirmed by this specification.
The weight of the constraint on the level of risk-sharing is significantly higher in the
case of large health and epizooty shocks. Additional communal losses of 1% due to the
H5N1 epizooty increases the propension of the enforcement constraint binding by 0.0086.
As health shocks are less correlated among sub-groups, the importance of the communal
effective exposure to health shocks should be less pronounced. The results confirm the
prediction of the model stating that idiosyncratic shocks are less exposed to potential
deviations of sub-groups. The importance of these shocks on the enforcement constraint
22
should be smaller than epizooty events where farmers without livestocks could represent a
deviating sub-group.
On the opposite, additional losses due to tropical typhoons loosen the enforcement
constraint. An additional marginal loss of 1% at communal level moves the marginal ratio
closer to the targeted ratio (perfect insurance conditional on the history) by 0.0064. The
more a commune is exposed to a natural disaster, the more efficient the risk-pooling will
be at commune level.
In the case of health and tropical typhoons, α is significantly higher than γ which confirms that a χ higher than the targeted ratio is accompanied by a higher weight given on
the constrained ratio for a commune affected by the average natural disaster shock (approximately 1% of the consumption). Thus, the difference between the weights attributed
to the targeted ratios when the enforcement is binding or not, would not be different21 from
0 for communities affected by a 10%-typhoon. For communities particularly affected by
tropical typhoons, we can not reject that enforcement constraints do not restrict instantaneous risk-sharing. On the other hand, the prediction that the sustainable contract should
be further to the targeted ratio with a higher χ is not verified in the case of epizooty shocks
for the average level of avian influenza exposure. For a 10%-avian influenza exposure, the
difference between the weight attributed to the targeted ratio is significant (11% with a
standard error of 3.6%).
Robustness checks
The identification method supposes that the surveyed households are part of the same
risk-pooling group. The estimation process will consequently underestimate the level of
risk-sharing in non-representative sub-groups by focusing only on the aggregate level. As
such, another interpretation of these results is that the group expands during a catastrophe
while epizooty or health shocks remain insured at a sub-group level.
Some of our assumptions are not crucial: it is possible to relax the hypothesis of constant
relative risk aversion22 but in a framework where households have similar expected income.
With heterogeneous households and large shocks, the hypothesis of constant relative risk
aversion (and thus the form of the utility function) is crucial. Our results could then
be driven by a particular form of risk aversion and not by an increase in community
ties. Nevertheless, as epizooty displays the same loss profile, the same results should be
observed if the effect was driven by an increasing relative loss aversion. Furthermore, the
results displayed in table 6 would suppose that relative risk aversion is decreasing in the
amplitude of the shocks. In other words, it would imply an increasing profile for the relative
risk aversion.
To control for inverse causality or for other cultural phenomena driving together effective exposure to tropical typhoons and informal transfers, we replicate the tests S1
presented above with the initial level of transfers (gifts and informal loans during the year
2003). As shown in table T1, the shocks are not correlated with past transfers. Similarly,
the estimation of the system S2 for natural disasters (table T3) proves that:
• the predictions of the model concerning enforcement issues with past transfers have
no influence on the weight attributed to the targeted ratio relatively to the placebo
ratio taking into account any potential amplitude of shocks. Past transfers are not
consistent with the model predictions.
21
for a 90% confidence interval.
with the logarithm specification, the exact value of this relative risk aversion does not influence the
estimations.
22
23
• the amplitude of natural disasters has no effect on the distance between the placebo
ratio and the targeted ratio, confirming the absence of a placebo effect driving our
results.
Another issue is the potential selection bias induced by the panel attrition. Households
(and consequently communes when we restrict our analysis to communes with 3 households) which disappear from the panel might precisely be those affected by a catastrophe
and excluded from informal risk-pooling groups. In a world where instantaneous risksharing is decided on frivolous parameters (a random draw for example), a household can
be temporary excluded from risk-sharing and thus not overcome the catastrophe and disappear from our database. In this case, natural disasters do not strengthen the community
links but "eliminate" households for which our measure of community link is temporary
low. Attrition issues might be mitigated by a couple of observations derived from the data:
communes losing households between 2004 and 2006 are not particularly affected by typhoons or different from the others by the level of initial informal transfers. Nevertheless,
these communes are more concerned by turnovers, but attrition is independent from the
combination of turnover and natural disasters.
Finally, the effect captured here could be explained by a greater response from internal
migrants in the wake of a typhoon having affected their relatives, rather than from the local
community. As explained earlier, the datasets do not disentangle presents from friends and
urban migrants. Informal gifts here encompass domestic remittances. As migrants have
no direct clue on the real level of income fluctuations of the household in normal times,
they might refrain themselves from compensating the household for unverifiable shocks.
The worst the disaster, the less uncertain the real level of losses might be. Improved
risk-pooling might then be explained by a loosened transparency constraint for migrants.
Two facts contribute to mitigate the importance of external assistance in our study: first,
considering successively the household as a unit, part of a commune, the commune as a
unit, part of a district, the districts and the provinces as units, parts of the entire Vietnam,
the only layer for which aggregate net gifts react to natural disaster shocks is the closest
to the nucleus. Transfers within district outside the communes and across districts are
not correlated with shocks at commune and district levels. Second, the elasticity of net
transfers to natural disaster shocks is the same wherever the household lies relatively to
the rest of the surveyed households in terms of income fluctuations. If migrants were to
insure the households against these shocks, the affected households would receive positive
net transfers but not supplied by the less affected households.
Risk-sharing through informal loans is more pronounced following natural disasters than
other shocks. The direct estimation of the theoretical model gives an empirical support
for the importance of enforcement constraints as limits to risk-pooling. The prevalence of
enforcement constraints for the group composed of the entire commune disappears following
heavy damages provoked by the passage of a typhoon. While a shock correlated among subgroups (epizooty) predicts tighter constraints in accordance with the theoretical model, the
amplitude of tropical typhoon reinforces the risk-pooling system at commune level. The
next section propose several tracks to identify precisely the mechanisms through which
risk-sharing is achieved at a higher layer than expected and described in the literature.
24
V.
Altruism and monitoring proxies (to be completed)
Risk-sharing at higher layer should certainly be privileged for co-moving shocks and responds to needs from potential participants in a risk-pooling group. That being said, severe
impediments related to the very nature of informal transfers are supposed to remain and
refrain agents from creating links between sub-groups of relatives, immediate neighbours
and friends. The community response to avian influenza confirms the prevalence of monitoring barriers and the limits to resource-pooling in groups with "distant monitoring". In
order to derive the reason why natural disasters allow monitoring issues to be side-stepped,
I investigate the importance of fertile grounds for coordinated communal response as determinants of risk-pooling. At the same time, I identify the factors influencing the individual
propensity to belong to the global risk-sharing group.
Recent movers’ monitoring capacities should be lower than those of settled families.
Similarly, the credibility of a threat exerted by the rest of a potential risk-sharing group
might be lower on new entrants and incorporating them might endanger the network sustainability. As a consequence, we would expect a smaller reliance on informal contracts
from households having settled in the village slightly before 2004. The table 7 confirms
that households having settled between 2000 and 2004 are excluded from risk-pooling in
the wake of a typhoon. It is not possible to reject that the correlation between individual
shocks and informal transfers is different from 0 for new entrants.
Table 7: Informal flows for recent movers following natural disasters
Specifications
Specification (S1)
OLS
Own shock × having moved recently
District FE
.372
(.197)†
-.387
(.084)∗∗
.608
(.245)∗
-.648
(.112)∗∗
.352
(.194)†
-.350
(.087)∗∗
.736
(.252)∗∗
-.693
(.122)∗∗
Controls for commune shocks
Sample
Yes
Total
Yes
Rural only
Yes
Total
Yes
Rural only
Observations
6508
4814
6508
4814
Own shock
Building on the previous results, we extend the analysis at commune-level. New entrants and households knowing that they will move in the next future represent a danger
for an established risk-sharing group. The model predicts a smaller level of risk-sharing
for communities where risk-sharing groups face uncertainty on their future composition.
Communes for which the turnover is high display lower risk pooling through informal loans
or donations. As shown in table 8, this effect at commune level is not completely supported
by surveyed households having moved for the past few years. Having newcomers as neighbors generates also less risk-pooling at commune level. The results seem to be essentially
explained by new entrants and not by departing households. They might be considered
as a simple extension of the previous table at commune level: the higher the turnover,
the higher the number of persons excluded from the extended group and the smaller the
reach of the extended structure. The proportion of temporary residency in the commune
25
does not influence the degree of risk-sharing. The status of temporary resident remains an
issue as it does not imply necessarily the presence of uncertainty in the composition of the
community.
Table 8: Informal flows following natural disasters depending on having moved or having
welcomed recent neighbors
Specifications
Specification (S1)
OLS
.634
(.249)∗
13.08
(5.96)∗
Own shock × new entrants
Own shock × turnover
.651
(.250)∗∗
District FE
.761
(.257)∗∗
11.20
(6.52)†
.773
(.257)∗∗
Own shock
-.598
(.118)∗∗
8.66
(4.50)†
-.629
(.115)∗∗
Yes
Yes
Yes
Yes
4662
4662
4662
4662
Observations
†
∗
-.654
(.128)∗∗
6.28
(4.63)
-.685
(.125)∗∗
∗∗
Significances are indicated at 10% , 5% , 1% . Communes for which we
have information on turnover are essentially rural.
In the same vein, if we follow Fafchamps & Gubert [2007], geographical distance attenuates the grip one household might have on the rest of the network. The table 9 illustrates
this idea since the greater the number of hamlets in a commune controlling for size effects,
the lower the level of risk-sharing. Similarly, cultural distance should matter as punishments might be linked to other activities (exclusion form new year celebrations...). In the
same table, we report the results from the basic regression with a dummy differentiating
households which belong to the local dominant ethnicity from the others. Controlling from
the local ethnicity and the ethnicity of the household, we find that households in a local
ethnic minority participate significantly less to risk-pooling in the aftermath of a typhoon.
These results replicate at commune level the results found by the literature at subgroup level. Nevertheless, it is not possible to conclude from this specification if additional
risk-sharing in united communes comes from a permanent higher level of cooperation or
from coordination in particularly bad times (table T6). Using the specification S2, we
bring support to the interpretation of the benchmark results presenting the constitution of
a consistent higher layer of risk-sharing as the main channel through which risk-pooling is
more efficient. This view is shared by Douty [1972] relying on anecdotal evidence: natural
disasters provoke the creation of a super-structure headed by pre-disaster leaders, enforcing
centralized transfers which would not be sustainable if decentralized. The effects present
for weak shocks in fractionalized communes tend to wash away for larger shocks. The figure
F3 illustrates this intuition. Communes with high level of fractionalization tend to catch up
with the other communes when the shock is larger. As soon as shocks represent more than a
certain threshold, the difference induced by the degree of fractionalization disappears. The
’double difference’ test in table T6 capture the amplitude of this catch-up effect. The results
are not completely similar for turnover. Turnover seems to matter whatever the amplitude
of the trauma. Above a certain threshold, it is not possible to reject the hypothesis that
turnover is harmless. Unfortunately, this result relies on exploding variance when the
26
Table 9: Informal flows following natural disasters depending on geographic dispersion and
being in an ethnic minority at commune level
Specifications
Specification (S1)
OLS
Own shock × geographic dispersion
3.22
(1.42)∗
Own shock × ethnic minority
Own shock
-.536
(.102)∗∗
Controls for size effects
Controls for the ethnicity
Yes
Yes
Observations
∗
2.83
(1.47)∗
.252
(.160)
-.882
(.179)∗∗
Yes
4662
†
District FE
-.540
(.111)∗∗
.357
(162)∗
-.932
(.185)∗∗
Yes
Yes
Yes
Yes
N
6508
4662
6508
Yes
∗∗
Significances are indicated at 10% , 5% , 1% . Communes for which we
have information on turnover are essentially rural.
trauma increases rather than on a clear "catch-up effect"23 . Unsurprisingly, turnover and
new entrants remain a stumbling block for establishing commune-level risk-sharing groups.
VI.
Influence of past shocks (to be completed)
The creation of a super-structure managing transfers between primary units (households)
is backed up by anecdotal evidence: in some communities affected regularly by dreadful
natural disasters, natural disaster funds centralize the transfers. Using past traumas, I test
for the presence of a learning-pattern in the ability to provide efficient risk-pooling after
the passage of a typhoon. First, I rely on recent shocks of the same type to see if these
lately affected communities present a higher level of resilience. Second, to depart from
the possibility that recent shocks might simply affect the expectations of the individuals,
I focus on other type of traumas and construct the exposure of each district to bombing
during the late years of the Vietnam war.
A.
Past natural disasters
In this section, I have computed the energy dissipated by 3 tropical typhoons (Eve, Wukong
and Kaemi) of the late 90’s at district-level. Unfortunately, the same precision for Thelma
(1997) is not available. As a consequence, I use the dissipated energy for the formers
and being close to the eye for the latter. The first results indicate that past exposure
could influence recent responses to catastrophes. The same regression considering assets’
transfers and formal instruments do not display the same learning pattern. The results do
not extend to idiosyncratic shock, indicating that this learning-pattern concerns specifically
the larger risk-pooling group designed to cover correlated fluctuations24 .
23
24
see table T6 for the ’double difference’ test and figure F2 for the illustration.
see table T4 in the appendix.
27
Table 10: Informal flows following natural disasters depending on past exposure
Specifications
Specification (S1)
OLS
Own shock × wind intensity (2000)
-2.13
(1.07)∗∗
Own shock × affected by Linda (1997)
Own shock
-.558
(.101)∗∗
Observations
District FE
-2.06
(1.24)∗∗
-.732
( .347)∗
-.481
(.102)∗∗
-.566
(.110)∗∗
Yes
6508
6508
-.761
(.347)∗
-.457
(.112)∗∗
Yes
6508
6508
Significances are indicated at 10%† , 5%∗ , 1%∗∗ . The wind intensity is normalized such
that a 10 points of percentage increase in energy dissipated relatively to the worst
outcome is associated with an increase of 21 points of percentage of risk-pooling.
An issue remains unchallenged: is this effect driven by a higher degree of cohesiveness
whatever the amplitude of the shock or is this effect driven by institutions especially builtup for coping with large natural disaster risks? As shown in table T5 and figure F4, the
second specification show an increasing profile for the influence of past shocks. Having
experienced the passage of a dreadful typhoons in the late 90’s does not increase the
community response to small shocks. However, the more a community is affected, the more
the experience of similar events matters. The ’double-difference’ test (see table T5 in the
appendix and figure F4 for the illustration) brings support to a divergence of risk-sharing
levels between experienced and unexperienced communes as the amplitude of the shock
rises. These results are consistent with anecdotal evidence; certain communes have indeed
institutionalized natural disaster funds in the Delta, responding to previous traumas. Such
coping mechanisms prove useful in exceptional situations. The next part of the article
builds on these empirical evidence and documents if increased ability to cope with natural
disasters is associated with specifically-designed institutions or a more global sentiment of
cohesiveness in a community.
B.
Bombing intensity between 1965 and 1975
VII.
Conclusion
28
References
Bellows, John, & Miguel, Edward. 2006. War and Institutions: New Evidence from
Sierra Leone. The American Economic Review, 96(2), 394–399.
Bellows, John, & Miguel, Edward. 2009. War and local collective action in Sierra
Leone. Journal of Public Economics, 93(11-12), 1144–1157.
Bloch, Francis, Genicot, Garance, & Ray, Debraj. 2007. Reciprocity in Groups
and the Limits to Social Capital. The American Economic Review, 97(2), 65–69.
Coate, S., & Ravallion, M. 1989. Reciprocity Without Commitment: Characterization and Performance of Informal Risk-Sharing Arrangements. Tech. rept. Warwick Development Economics Research Centre.
Douty, Christopher M. 1972. Disasters and Charity: Some Aspects of Cooperative
Economic Behavior. The American Economic Review, 62(4), 580–590.
Eisensee, Thomas, & Strömberg, David. 2007. News Droughts, News Floods, and
U.S. Disaster Relief. Quarterly Journal of Economics, 122(2), 693–728.
Fafchamps, Marcel, & Gubert, Flore. 2007. The formation of risk sharing networks.
Journal of Development Economics, 83(2), 326–350.
Fafchamps, Marcel, & Lund, Susan. 2003. Risk-sharing networks in rural Philippines.
71(Aug.), 261–287.
Foster, Andrew D., & Rosenzweig, Mark R. 2001. Imperfect Commitment, Altruism, and the Family: Evidence from Transfer Behavior in Low-Income Rural Areas. The
Review of Economics and Statistics, 83(3), 389–407.
Genicot, Garance, & Ray, Debraj. 2003. Group Formation in Risk-Sharing Arrangements. The Review of Economic Studies, 70(1), 87–113.
Glaeser, Edward L., Laibson, David, & Sacerdote, Bruce. 2002. An Economic
Approach to Social Capital. The Economic Journal, 112(483), F437–F458.
Jacoby, Hanan G. 1993. Shadow Wages and Peasant Family Labour Supply: An Econometric Application to the Peruvian Sierra. The Review of Economic Studies, 60(4),
903–921.
Kochar, Anjini. 1999. Smoothing Consumption by Smoothing Income: Hours-of-Work
Responses to Idiosyncratic Agricultural Shocks in Rural India. Review of Economics and
Statistics, 81(1), 50–61.
Ligon, Ethan, Thomas, Jonathan P., & Worrall, Tim. 2002. Informal Insurance Arrangements with Limited Commitment: Theory and Evidence from Village
Economies. The Review of Economic Studies, 69(1), 209–244.
Paxson, Christina H. 1992. Using Weather Variability to Estimate the Response of
Savings to Transitory Income in Thailand. The American Economic Review, 82(1),
15–33.
Pollak, Robert A. 1976. Interdependent Preferences. The American Economic Review,
66(3), 309–320.
29
Rosenzweig, Mark R. 1988. Risk, Implicit Contracts and the Family in Rural Areas of
Low-Income Countries. The Economic Journal, 98(393), 1148–1170.
Yang, Dean, & Choi, HwaJung. 2007. Are Remittances Insurance? Evidence from
Rainfall Shocks in the Philippines. World Bank Econ Rev, May, lhm003.
30
A
Intuition behind the model
Figure F1: The optimal contract and the evolution of marginal ratios (ζ=λ̄)
1. A shock affects the individual 2, without binding the constraint for the individual 1.
2. Transfers from 1 to 2 are refrained so that deviation is not preferable for 1.
3. A shock affects the individual 1, without binding the constraint for the individual 2.
4. Transfers from 2 to 1 are refrained so that deviation is not preferable for 2.
5. An increase in punishment threat allows the households to keep the ratio constant.
31
B
Robustness checks
Table T1: Informal flows following natural disasters using past transfers
Specification (S1)
Informal loans
Coeff. (SE)
Own shock
-.036
-.019
District fixed effects
Observations
(.055)
(.074)
Yes
4814
Gifts
Coeff. (SE)
.044
-.013
(.058)
(.077)
Yes
4814
Significances are indicated at 10%† , 5%∗ , 1%∗∗ . The sample
is restricted on rural areas. Results are similar with the full
sample.
Table T2: Informal flows following epizooty shocks depending on turnover at communelevel
Specifications
Specification (S1)
OLS
Own shock × new entrants
.215
(.143)
.238
(1.05)
Own shock × turnover
.205
(.144)
Own shock
-.019
(0.051)
2.15
(1.68)
-.013
(.051)
Observations
Yes
4870
Yes
4870
District FE
.163
(.147)
.194
(1.06)
.154
(.148)
-.064
(.053)
1.92
(1.70)
-.058
(.053)
Yes
4870
Yes
4870
Significances are indicated at 10%† , 5%∗ , 1%∗∗ . Communes for which we have information on turnover are essentially rural.
32
Table T3: Informal flows following natural disasters using past transfers
Specification (S2)
Type of shocks
Natural shocks
Specifications
Overweight induced
Targeted ratio
Constrained ratio
Co. OLS Un. OLS
Binding constraint
.154
.402
(.336)
(.334)
.347
.121
(.023)
(.025)
.655
.471
(.020)
(.024)
Non-binding constraint
.348
.079
(.022)
(.023)
.652
.446
(.022)
(.023)
Targeted ratio
Constrained ratio
Targeted ratios
Constrained ratios
Observations
Tests
-.001
(.030)
.001
(.030)
1766
for equality
.024
(.033)
.042
(.034)
1665
Significances are indicated for the important variables
at 10%† , 5%∗ , 1%∗∗ .
Table T4: Robustness checks for past exposure specification
Specification (S1)
Health shocks
Type of shocks
Type of transfers
Own shock × wind intensity (2000)
Own shock × affected by Linda (1997)
Own shock
Informal transfers
-.030
(.074)
.033
(.029)
-.016
-.018
(.006)∗ (.005)∗∗
Transfers of assets
-.710
(5.12)
.676
( 1.42)
-.670
-.700
(.453) (.460)
Yes
Yes
Yes
Yes
District fixed effects
Observations
6513
†
∗
Natural disasters
∗∗
6513
6508
6508
Significances are indicated at 10% , 5% , 1% . Coefficients and standard errors for
wind intensity are multiplied by 105 for readability concerns.
33
C
Additional results using specification (S2)
Table T5: Informal flows following natural disasters depending on past exposure
Specification (S2)
Natural shocks
Type of shocks
Specifications
Overweight induced (shock × past exposure)
Overweight induced (Past exposure)
Controls for overweight
ML (FE) ML
Results
.891
2.68
(.434)∗
(.961)∗∗
.001
(.019)
Yes
-.041
(.047)
Yes
Overweight - 2% shock
.032
.151
(.045)
(.107)
.004
-.027
(.009)
(.019)
For an experienced commune
For a unexperienced commune
Overweight
.177
(.097)†
.020
(.047)
For an experienced commune
For a unexperienced commune
Difference exp/unexp. & 10/2%
Provinces-fixed effects
District-fixed effects
Observations
- 10% shock
.428
(.239)†
-.129
(.100)
Double-difference tests
.142
.430
(.069)∗
(.153)∗∗
Yes
1756
Yes
1756
Significances are indicated for the important variables at 10%† , 5%∗ , 1%∗∗ .
An experienced commune has been hit by at least one of the dreadful
typhoons at the end of the 90’s.
34
Table T6: Informal flows following natural disasters depending on geographic dispersion
and being in an ethnic minority at commune level
Specification (S2)
Natural shocks
Type of shocks
Specifications
Overweight induced (shock × ethnic minority)
Overweight induced (shock × turnover)
Controls for overweights (shock, turnover, minority)
ML (FE) ML (FE)
Results
6.63
(3.38)∗
-.889
(13.3)
Yes
Yes
-.003
(.065)
.058
(.023)∗∗
-.632
(.311)∗
.012
(.015)
Fractionalized commune
United commune
High turnover
No turnover
.497
(.258)∗
.293
(.119)∗
-.616
(.799)
.063
(.078)
Fractionalized commune
United commune
High turnover
No turnover
Difference high/low frac. & 10/2%
Difference high/low turn. & 10/2%
Controls for ethnicity
Provinces-fixed effects
Observations
Double-difference tests
.265
(.135)∗
.035
(.534)
Yes
1766
Yes
1665
Significances are indicated for the important variables at 10%† , 5%∗ , 1%∗∗ . Controls
for ethnicity include individual and commune dummies. High turnover correspond
to a 50% renewal of the commune in 5 years. A fractionalized commune is here
composed of a main ethnic group representing half of the surveyed households. A
united commune is composed of a main ethnic group representing all the surveyed
households.
35
Figure F2: Overweight as a function of the amplitude of the shock and turnover at commune
level
Figure F3: Overweight as a function of the amplitude of the shock and fractionalization at
commune level
36
Figure F4: Overweight as a function of the amplitude of the shock and experience
37
Figure F5: Informal transfer intensity in each district using the household survey in 2004
38
Figure F6: Informal transfer intensity in each district using the household survey in 2006
39

Do tropical typhoons smash community ties? Theory and Evidence

Transcription

Similar documents

No Drill Modern Gas Shock Installation on a 1931 Ford

Brochures - Bull Realty

Circular Flow Chart

READy-TO-RUN DISTRIBUTOR From Professional Products comes

Ken Goodgames Chief Executive Officer Transformance

Pureflow Water Solutions

Transition Black Isle

Miami Towne Center.p65

File